-
Notifications
You must be signed in to change notification settings - Fork 425
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Rework convolution nets (CNN) in Java and C# #31
Labels
Comments
jeffheaton
added a commit
that referenced
this issue
Jan 14, 2016
jeffheaton
added a commit
that referenced
this issue
Jan 18, 2016
jeffheaton
added a commit
that referenced
this issue
Jan 23, 2016
… likely debugging needed. Convergance rates could be better. #31
Java examples have been reworked to use DL4J. Next the C# will be updated for CNTK. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Right now the CNN's attempt to use vectorized data structures between the layers, not the 3D "volume" type structure common in CNN's. While this does ensure a great deal of compatibility with the 1D vectors commonly used in non-CNN neural networks, it makes the code difficult to understand and debug. There seems to be a bug that is currently having a bad effect on the accuracy of the CNN's in C#/Java. Rather than attempt to get the vectorized version working, I am going to convert to the "volume" design. This will be more compatible with the Theano Python implementation. I can also minimize disruption to the non-CNN parts.
Hope to have this change in before the end of January (at the latest). Until then, the Java/C# CNN's do not work well.
The text was updated successfully, but these errors were encountered: